Decoding and mapping task states of the human brain via deep learning.

Area of Science:

• Neuroscience
• Machine Learning
• Neuroimaging

Background:

• Functional magnetic resonance imaging (fMRI) is crucial for understanding brain activity.
• Multivariate pattern analysis (MVPA) with Support Vector Machines (SVM) is common for decoding brain states but requires expert-defined features.
• Feature engineering in SVM-MVPA can be labor-intensive and limit generalizability.

Purpose of the Study:

• To introduce a Deep Neural Network (DNN) for direct decoding of multiple brain task states from fMRI data.
• To eliminate the need for manual feature selection in brain state decoding.
• To evaluate the DNN's performance and generalizability, including transfer learning capabilities.

Main Methods:

• A DNN classifier was trained and tested on task fMRI data from the Human Connectome Project (N=1,034).
• The DNN was evaluated for its ability to decode seven distinct tasks.
• Transfer learning performance was assessed on smaller datasets (N=43) for working memory and motor tasks, comparing against SVM-MVPA.

Main Results:

• The DNN achieved an average accuracy of 93.7% in identifying seven brain tasks.
• For transfer learning on smaller datasets, the DNN reached 89.0% (working memory) and 94.7% (motor task) accuracy.
• These accuracies significantly surpassed SVM-MVPA results (69.2% and 68.6% respectively).
• Network visualization confirmed the DNN automatically identified task-relevant brain regions.

Conclusions:

• The proposed DNN method accurately decodes brain task states from fMRI data without manual feature engineering.
• The DNN demonstrates superior performance and generalizability, especially for transfer learning on limited datasets.
• This deep decoding approach offers a powerful and efficient alternative for fMRI research.